Master thesis : Development of a robot interface for cognitive experimental tasks in human functional magnetic resonance imaging

Abstract

This work presents the project that I realized during my four months internship inside the Courtois NeuroMod project, in Montreal, from February to June 2022. The Courtois project on neuronal modeling (CNeuroMod) aims to collect 500h of functional neuroimaging data per subject, on 6 subjects over a period of 5 years using a range of natural stimuli. This large multimodal data set is then used to train artificial neural networks that mimic human behavior and brain activity in multiple tasks, following a brain-augmented learning paradigm. One of the future projects of the lab is to explore the field of robotics and, more specifically, to embed brain-augmented models in small robots, trained in real-world conditions. The first part of this project presents the development of an Application Programming Interface (API) to control a commercial robot, called the Cozmo robot. The final goal is to be able to control the robot from inside a Magnetic Resonance Imaging (MRI) scanner with a video game controller. Since the lab develops deep learning architectures for Reinforcement Learning (RL) tasks, the developed API also provides a framework for RL with the robot. The API is based on an existing interface, and is designed to provide a simple user interface. This API is then used to deploy a control task in a MRI machine, by integrating the API into the software framework already in place for the experiments within the NeuroMod project. As the robot can be located anywhere in the building, the implementation manages data transfers between machines so that the control can be done in real time and data of interest can be retrieved in addition to those acquired by the MRI. A human subject is finally able to control a remote robot using a joystick, from inside an MRI machine acquiring its brain data. An automatic tracking system of the robot is also developed.